, 2025) and developed a three-stage training pipeline.
Video world model uses test-time training as persistent memory for long-range consistency
World-T3 consolidates past observations into parametric memory via online recursive least-squares, enabling consistent generation with constant storage and computational cost.
Academic
Junyi Wu · Zekun Li · Chenhongyi Yang · Chen Guo · Lingchen Yang · Yongan Zhang · +6 more
Meta Reality Labs · Brown University · University of Illinois Chicago
Research Digest··3 min read
The authors introduce World-T3, a camera-controlled video world model that maintains persistent scene memory through test-time training (TTT).
Why this paper
From Meta Reality Labs and 2 others
In one line
A camera-controlled video world model uses test-time training memory to preserve observed scenes in fixed-size parameters, improving long-horizon revisit consistency.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (hardware single NVIDIA B200 GPU)
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§